Overview
- The 2026 AI Index, released Monday by Stanford HAI, finds mass use of generative AI alongside a rise in real‑world harms, with population adoption near 53% and documented incidents climbing to 362 in 2025.
- Developers are sharing less about how systems are built or behave, as the Foundation Model Transparency Index fell from 58 in 2024 to 40 in 2025 and access shifted to API‑only releases that restrict outside checks.
- China has largely closed the performance gap on top models, even as the U.S. retains a dominant edge in private investment at $285.9 billion versus $12.4 billion and operates vast data‑center capacity measured at 29.6 gigawatts.
- Studies show solid gains on narrow tasks such as customer support and coding, yet macro impact remains small and early‑career jobs are taking the strain, with U.S. software developer employment ages 22–25 down about 20% from its 2022 peak.
- AI’s energy and water demands are mounting, with one training run for Grok 4 estimated at 72,816 tons of CO2 equivalent and GPT‑4o inference projected to use water equal to the drinking needs of about 12 million people, fueling local pushback that has delayed or blocked billions in U.S. data‑center projects.